161 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 52; RUN: opt < %s -passes=amdgpu-sw-lower-lds -S -mtriple=amdgcn-amd-amdhsa | FileCheck %s3 4; Test to check if static LDS accesses in kernel are lowered correctly. Also checks if amdgpu-no-heap-ptr attribute5; is removed from the list of attributes6@lds_1 = internal addrspace(3) global [1 x i8] poison, align 47@lds_2 = internal addrspace(3) global [1 x i32] poison, align 88 9;.10; CHECK: @llvm.amdgcn.sw.lds.k0 = internal addrspace(3) global ptr poison, no_sanitize_address, align 8, !absolute_symbol [[META0:![0-9]+]]11; CHECK: @llvm.amdgcn.sw.lds.k0.md = internal addrspace(1) global %llvm.amdgcn.sw.lds.k0.md.type { %llvm.amdgcn.sw.lds.k0.md.item { i32 0, i32 8, i32 32 }, %llvm.amdgcn.sw.lds.k0.md.item { i32 32, i32 1, i32 32 }, %llvm.amdgcn.sw.lds.k0.md.item { i32 64, i32 4, i32 32 } }, no_sanitize_address12;.13define amdgpu_kernel void @k0() sanitize_address #1 {14; CHECK-LABEL: define amdgpu_kernel void @k0(15; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {16; CHECK-NEXT: [[WID:.*]]:17; CHECK-NEXT: [[TMP0:%.*]] = call i32 @llvm.amdgcn.workitem.id.x()18; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.amdgcn.workitem.id.y()19; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.amdgcn.workitem.id.z()20; CHECK-NEXT: [[TMP3:%.*]] = or i32 [[TMP0]], [[TMP1]]21; CHECK-NEXT: [[TMP4:%.*]] = or i32 [[TMP3]], [[TMP2]]22; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i32 [[TMP4]], 023; CHECK-NEXT: br i1 [[TMP5]], label %[[MALLOC:.*]], label %[[BB20:.*]]24; CHECK: [[MALLOC]]:25; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE:%.*]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 2, i32 0), align 426; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 2, i32 2), align 427; CHECK-NEXT: [[TMP8:%.*]] = add i32 [[TMP6]], [[TMP7]]28; CHECK-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i6429; CHECK-NEXT: [[TMP10:%.*]] = call ptr @llvm.returnaddress(i32 0)30; CHECK-NEXT: [[TMP11:%.*]] = ptrtoint ptr [[TMP10]] to i6431; CHECK-NEXT: [[TMP12:%.*]] = call i64 @__asan_malloc_impl(i64 [[TMP9]], i64 [[TMP11]])32; CHECK-NEXT: [[TMP13:%.*]] = inttoptr i64 [[TMP12]] to ptr addrspace(1)33; CHECK-NEXT: store ptr addrspace(1) [[TMP13]], ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, align 834; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i64 835; CHECK-NEXT: [[TMP15:%.*]] = ptrtoint ptr addrspace(1) [[TMP14]] to i6436; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP15]], i64 24)37; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i64 3338; CHECK-NEXT: [[TMP17:%.*]] = ptrtoint ptr addrspace(1) [[TMP16]] to i6439; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP17]], i64 31)40; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i64 6841; CHECK-NEXT: [[TMP19:%.*]] = ptrtoint ptr addrspace(1) [[TMP18]] to i6442; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP19]], i64 28)43; CHECK-NEXT: br label %[[BB20]]44; CHECK: [[BB20]]:45; CHECK-NEXT: [[XYZCOND:%.*]] = phi i1 [ false, %[[WID]] ], [ true, %[[MALLOC]] ]46; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()47; CHECK-NEXT: [[TMP21:%.*]] = load ptr addrspace(1), ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, align 848; CHECK-NEXT: [[TMP22:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 1, i32 0), align 449; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds i8, ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, i32 [[TMP22]]50; CHECK-NEXT: [[TMP24:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 2, i32 0), align 451; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i8, ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, i32 [[TMP24]]52; CHECK-NEXT: [[TMP26:%.*]] = ptrtoint ptr addrspace(3) [[TMP23]] to i3253; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP26]]54; CHECK-NEXT: [[TMP28:%.*]] = ptrtoint ptr addrspace(1) [[TMP27]] to i6455; CHECK-NEXT: [[TMP29:%.*]] = lshr i64 [[TMP28]], 356; CHECK-NEXT: [[TMP30:%.*]] = add i64 [[TMP29]], 214745088057; CHECK-NEXT: [[TMP31:%.*]] = inttoptr i64 [[TMP30]] to ptr58; CHECK-NEXT: [[TMP32:%.*]] = load i8, ptr [[TMP31]], align 159; CHECK-NEXT: [[TMP33:%.*]] = icmp ne i8 [[TMP32]], 060; CHECK-NEXT: [[TMP34:%.*]] = and i64 [[TMP28]], 761; CHECK-NEXT: [[TMP35:%.*]] = trunc i64 [[TMP34]] to i862; CHECK-NEXT: [[TMP36:%.*]] = icmp sge i8 [[TMP35]], [[TMP32]]63; CHECK-NEXT: [[TMP37:%.*]] = and i1 [[TMP33]], [[TMP36]]64; CHECK-NEXT: [[TMP38:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 [[TMP37]])65; CHECK-NEXT: [[TMP39:%.*]] = icmp ne i64 [[TMP38]], 066; CHECK-NEXT: br i1 [[TMP39]], label %[[ASAN_REPORT:.*]], label %[[BB42:.*]], !prof [[PROF2:![0-9]+]]67; CHECK: [[ASAN_REPORT]]:68; CHECK-NEXT: br i1 [[TMP37]], label %[[BB40:.*]], label %[[BB41:.*]]69; CHECK: [[BB40]]:70; CHECK-NEXT: call void @__asan_report_store1(i64 [[TMP28]]) #[[ATTR6:[0-9]+]]71; CHECK-NEXT: call void @llvm.amdgcn.unreachable()72; CHECK-NEXT: br label %[[BB41]]73; CHECK: [[BB41]]:74; CHECK-NEXT: br label %[[BB42]]75; CHECK: [[BB42]]:76; CHECK-NEXT: store i8 7, ptr addrspace(1) [[TMP27]], align 477; CHECK-NEXT: [[TMP43:%.*]] = ptrtoint ptr addrspace(3) [[TMP25]] to i3278; CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP43]]79; CHECK-NEXT: [[TMP45:%.*]] = ptrtoint ptr addrspace(1) [[TMP44]] to i6480; CHECK-NEXT: [[TMP46:%.*]] = add i64 [[TMP45]], 381; CHECK-NEXT: [[TMP47:%.*]] = inttoptr i64 [[TMP46]] to ptr addrspace(1)82; CHECK-NEXT: [[TMP48:%.*]] = ptrtoint ptr addrspace(1) [[TMP44]] to i6483; CHECK-NEXT: [[TMP49:%.*]] = lshr i64 [[TMP48]], 384; CHECK-NEXT: [[TMP50:%.*]] = add i64 [[TMP49]], 214745088085; CHECK-NEXT: [[TMP51:%.*]] = inttoptr i64 [[TMP50]] to ptr86; CHECK-NEXT: [[TMP52:%.*]] = load i8, ptr [[TMP51]], align 187; CHECK-NEXT: [[TMP53:%.*]] = icmp ne i8 [[TMP52]], 088; CHECK-NEXT: [[TMP54:%.*]] = and i64 [[TMP48]], 789; CHECK-NEXT: [[TMP55:%.*]] = trunc i64 [[TMP54]] to i890; CHECK-NEXT: [[TMP56:%.*]] = icmp sge i8 [[TMP55]], [[TMP52]]91; CHECK-NEXT: [[TMP57:%.*]] = and i1 [[TMP53]], [[TMP56]]92; CHECK-NEXT: [[TMP58:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 [[TMP57]])93; CHECK-NEXT: [[TMP59:%.*]] = icmp ne i64 [[TMP58]], 094; CHECK-NEXT: br i1 [[TMP59]], label %[[ASAN_REPORT1:.*]], label %[[BB62:.*]], !prof [[PROF2]]95; CHECK: [[ASAN_REPORT1]]:96; CHECK-NEXT: br i1 [[TMP57]], label %[[BB60:.*]], label %[[BB61:.*]]97; CHECK: [[BB60]]:98; CHECK-NEXT: call void @__asan_report_store1(i64 [[TMP48]]) #[[ATTR6]]99; CHECK-NEXT: call void @llvm.amdgcn.unreachable()100; CHECK-NEXT: br label %[[BB61]]101; CHECK: [[BB61]]:102; CHECK-NEXT: br label %[[BB62]]103; CHECK: [[BB62]]:104; CHECK-NEXT: [[TMP63:%.*]] = ptrtoint ptr addrspace(1) [[TMP47]] to i64105; CHECK-NEXT: [[TMP64:%.*]] = lshr i64 [[TMP63]], 3106; CHECK-NEXT: [[TMP65:%.*]] = add i64 [[TMP64]], 2147450880107; CHECK-NEXT: [[TMP66:%.*]] = inttoptr i64 [[TMP65]] to ptr108; CHECK-NEXT: [[TMP67:%.*]] = load i8, ptr [[TMP66]], align 1109; CHECK-NEXT: [[TMP68:%.*]] = icmp ne i8 [[TMP67]], 0110; CHECK-NEXT: [[TMP69:%.*]] = and i64 [[TMP63]], 7111; CHECK-NEXT: [[TMP70:%.*]] = trunc i64 [[TMP69]] to i8112; CHECK-NEXT: [[TMP71:%.*]] = icmp sge i8 [[TMP70]], [[TMP67]]113; CHECK-NEXT: [[TMP72:%.*]] = and i1 [[TMP68]], [[TMP71]]114; CHECK-NEXT: [[TMP73:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 [[TMP72]])115; CHECK-NEXT: [[TMP74:%.*]] = icmp ne i64 [[TMP73]], 0116; CHECK-NEXT: br i1 [[TMP74]], label %[[ASAN_REPORT2:.*]], label %[[BB77:.*]], !prof [[PROF2]]117; CHECK: [[ASAN_REPORT2]]:118; CHECK-NEXT: br i1 [[TMP72]], label %[[BB75:.*]], label %[[BB76:.*]]119; CHECK: [[BB75]]:120; CHECK-NEXT: call void @__asan_report_store1(i64 [[TMP63]]) #[[ATTR6]]121; CHECK-NEXT: call void @llvm.amdgcn.unreachable()122; CHECK-NEXT: br label %[[BB76]]123; CHECK: [[BB76]]:124; CHECK-NEXT: br label %[[BB77]]125; CHECK: [[BB77]]:126; CHECK-NEXT: store i32 8, ptr addrspace(1) [[TMP44]], align 2127; CHECK-NEXT: br label %[[CONDFREE:.*]]128; CHECK: [[CONDFREE]]:129; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()130; CHECK-NEXT: br i1 [[XYZCOND]], label %[[FREE:.*]], label %[[END:.*]]131; CHECK: [[FREE]]:132; CHECK-NEXT: [[TMP78:%.*]] = call ptr @llvm.returnaddress(i32 0)133; CHECK-NEXT: [[TMP79:%.*]] = ptrtoint ptr [[TMP78]] to i64134; CHECK-NEXT: [[TMP80:%.*]] = ptrtoint ptr addrspace(1) [[TMP21]] to i64135; CHECK-NEXT: call void @__asan_free_impl(i64 [[TMP80]], i64 [[TMP79]])136; CHECK-NEXT: br label %[[END]]137; CHECK: [[END]]:138; CHECK-NEXT: ret void139;140 store i8 7, ptr addrspace(3) @lds_1, align 4141 store i32 8, ptr addrspace(3) @lds_2, align 2142 ret void143}144 145!llvm.module.flags = !{!0}146!0 = !{i32 4, !"nosanitize_address", i32 1}147attributes #1 = { "amdgpu-no-heap-ptr" }148;.149; CHECK: attributes #[[ATTR0]] = { sanitize_address "amdgpu-lds-size"="8" }150; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }151; CHECK: attributes #[[ATTR2:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) }152; CHECK: attributes #[[ATTR3:[0-9]+]] = { convergent nocallback nofree nounwind willreturn }153; CHECK: attributes #[[ATTR4:[0-9]+]] = { convergent nocallback nofree nounwind willreturn memory(none) }154; CHECK: attributes #[[ATTR5:[0-9]+]] = { convergent nocallback nofree nounwind }155; CHECK: attributes #[[ATTR6]] = { nomerge }156;.157; CHECK: [[META0]] = !{i32 0, i32 1}158; CHECK: [[META1:![0-9]+]] = !{i32 4, !"nosanitize_address", i32 1}159; CHECK: [[PROF2]] = !{!"branch_weights", i32 1, i32 1048575}160;.161